Profile cover photo
Profile photo
Mihai Oltean
14 followers
14 followers
About
Mihai's posts

Post has attachment

Post has attachment

Post has attachment
LIDAR map created with Tera Ranger One sensor


Photo

Post has attachment

New legs for the Jenny 5 robot for better stability. This is made from a double set of aluminium profiles.

Source is already on the github: https://github.com/jenny5-robot

Photo

Source code for Traceless Genetic Programming is now on GitHub

Traceless Genetic Programming (TGP) is a very fast variant of standard Genetic Programming (GP) aiming at solving regression and classification problems. The big difference between TGP and GP is that TGP does not store the computer programs, instead it stores only the output of a computer program for each training / validation / test data. Because TGP does not recompute the entire tree, it is several times faster than standard GP.

Now I have improved the original source code for solving even-parity and classification problems and I have posted on GitHub:

https://github.com/mihaioltean/genetic-programming

Post has attachment
New tracks for robot

#jenny5robot

I just added new tracks for the Jenny 5 robot. These new tracks are intended to better overcome small rocks / obstacles.

Here is a picture of them.

CAD files can be downloaded from GitHub repository:

https://github.com/jenny5-robot/jenny5-cad
Photo

Genetic algorithm with binary encoding for function optimization

C++ implementation for a Genetic Algorithm with binary encoding for function optimization is available from: https://github.com/mihaioltean/genetic-algorithms

Rastrigin function is implemented as an example of multimodal test function.

Post has attachment
Automatically Defined Functions in Multi Expression Programming

I just added a C++ implementation of MEP with ADFs to my GitHub repository: https://github.com/mepx/mep-basic-src

Multi Expression Programming (MEP) is a Genetic Programming variant which uses a linear representation of computer programs. MEP has the ability to encode multiple solutions in the same chromosome, thus exploring the search space better than Genetic Programming techniques encoding a single solution in a chromosome.

With Automatically Defined Functions (ADFs), MEP is able to solve larger instances than without ADFs. I have added a test example for designing a digital circuit for the even-parity problem with 6 inputs. Without ADFs, Genetic Programming is not able to solve it using reasonable resources (population size and time). 

Post has attachment
Jenny 5 CAD files now available

#jenny5robot

I just uploaded on GitHub the first version of the CAD files for the Jenny 5 robot:

https://github.com/jenny5-robot/jenny5-cad

You need OpenSCAD (www.openscad.org) to view / modify the files.

Go to folder "robot" and load robot.scad and then press F5. It will take few seconds (or minutes if you have a slow computer) to render.

This is not the final version! A lot of details are left out.
Photo

Post has attachment
Tera Ranger One library

#jenny5robot

For obstacles avoidance in Jenny 5 robot (www.jenny5.org) I have decided to use one (or more) LIDAR sensor(s). After searching the internet for a light and good sensor I have decided to go ahead with Tera Ranger One:

www.teraranger.com/products/teraranger-one/

It has long range (up to 14 meters - which is more than I ever need for this robot) and a decent precision (4cm error). The user can connect in multiple ways to this sensor, one of them being USB.

For reading data generated by this sensor I have written a C++ library which can be downloaded from:

https://github.com/mihaioltean/tera-ranger-one-lib

The library is asynchronous, so that it will not block the program while waiting for sensor to respond.

The library has a MIT license, so that anyone could use it for any purpose.
Photo
Wait while more posts are being loaded